Manifesto

Robots are moving from labs into the real world: warehouses, homes, farms, cities, but there's a bottleneck nobody talks about.

Software ships fast because every change is tested automatically, push code, run tests, and know what broke in seconds. Decades of tooling make this possible, from CI/CD and test suites to monitoring and observability.

Robots have none of this.

You train a policy, run it on a robot, and it fails. So you SSH in, pull the logs, download the video, and scrub through camera footage trying to figure out what went wrong, one run at a time, hundreds of runs per week, with no patterns detected and no memory between sessions. Every session starts from scratch.

This is the iteration loop for every robotics team on the planet, and it's slow.

The bottleneck isn't training models; GPUs are fast. It's everything after: testing every change before it ships, and understanding what happened when it fails, at scale, across thousands of runs, automatically.

We believe this is a solvable problem.

We're building continuous evaluation for robotics: a control plane that turns every policy or stack change into evaluation runs. It selects the cases that matter and dispatches them to managed runners: simulation that stresses policies across seeds and perturbations, and field replay that turns real incidents into regression tests. Every change comes back as one verdict, pass or fail, with the evidence attached.

When a check fails, the platform doesn't hand you a red X — it hands you evidence. Subtasks and behavior are annotated on the run and lined up with sensor anomalies, so the failed window points to the problem: data, policy, or hardware. The engineer knows their system. We give them the evidence to decide fast.

And every run compounds. Results become part of the release record, failed windows become curated training data, and every failure becomes a new case in the suite. The test suite grows from exactly what broke in the field — which is how software teams got fast, and how robotics teams will.

This is a hard problem. Evaluating physical behavior across robot types, sensor modalities, and environments is unsolved. We're not afraid of that. Hard problems worth solving are the only ones that matter.

If we succeed, robots iterate as fast as software. And when robots iterate fast, they deploy everywhere and diffuse into every part of society.